Google DeepMind Released an AI Model that Simulates How a Fruit Fly Behaves

Google DeepMind, in collaboration with the Howard Hughes Medical Institute’s Janelia Research Campus, has developed an advanced AI-powered model that accurately simulates the behavior, movement, and flight of a fruit fly.

This lifelike virtual insect is part of a broader initiative to use AI and simulation to understand animal behavior at an unprecedented level of detail.

The project leverages MuJoCo—DeepMind’s open-source physics simulator originally designed for robotics and biomechanics—which has been enhanced with new capabilities to replicate the complex physics of insect locomotion and flight.

These include features like fluid force simulation for wing flapping and adhesion actuators to mimic the gripping ability of insect feet.

To breathe life into the virtual fly, DeepMind trained an artificial neural network on real-world behavioral footage of fruit flies.

This network now controls the digital insect within MuJoCo, enabling it to exhibit strikingly realistic motion and even use visual cues to guide its actions.

The goal of the project is to provide scientists with tools to study the connections between brain activity, bodily movement, and environmental interaction—connections that are often difficult to measure in living organisms.

DeepMind has previously applied this method to simulate rodent behavior and is now preparing to create a virtual zebrafish model, a species that shares around 70% of its protein-coding genes with humans.

To support the wider scientific community, DeepMind and Janelia have open-sourced the fruit fly model, enabling researchers around the world to explore and expand on this cutting-edge work.